Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75033.67646950291,
-      "required_pages_per_second_upper_estimate": 0.005191003840276211
+      "required_token_positions_per_second": 76705.62071556476,
+      "required_pages_per_second_upper_estimate": 0.005306672822663904
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9379.209558687864,
-      "required_pages_per_second_upper_estimate": 0.0039268791820550215
+      "required_token_positions_per_second": 9588.202589445595,
+      "required_pages_per_second_upper_estimate": 0.004014380199762523
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18758.419117375728,
-      "required_pages_per_second_upper_estimate": 0.0039068840093705905
+      "required_token_positions_per_second": 19176.40517889119,
+      "required_pages_per_second_upper_estimate": 0.003993939482950552
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28137.628676063596,
-      "required_pages_per_second_upper_estimate": 0.0016868078754587014
+      "required_token_positions_per_second": 28764.607768336788,
+      "required_pages_per_second_upper_estimate": 0.0017243943146988373
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28137.628676063596,
-      "required_pages_per_second_upper_estimate": 0.002187576865215218
+      "required_token_positions_per_second": 28764.607768336788,
+      "required_pages_per_second_upper_estimate": 0.002236321731849884
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28137.628676063596,
-      "required_pages_per_second_upper_estimate": 0.36091928882471486
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28764.607768336788,
+      "required_pages_per_second_upper_estimate": 0.3689615034226958
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.11158120001539373,
-    "producer_p95_seconds": 0.30687986399425426,
-    "gpu_update_seconds": 0.9825987938891199,
-    "producer_batches_per_second": 8.962083217083524,
-    "gpu_batches_per_second": 1.0177093705173463,
-    "producer_headroom": 8.806132159840194,
+    "producer_batch_seconds": 0.27941910324204855,
+    "producer_p95_seconds": 0.306843584003218,
+    "gpu_update_seconds": 0.9611811926194275,
+    "producer_batches_per_second": 3.578853372576118,
+    "gpu_batches_per_second": 1.0403865656950515,
+    "producer_headroom": 3.4399265528627736,
     "sustainable": true,
-    "buffer_seconds": 60.92112522112543,
+    "buffer_seconds": 60.55441513502393,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 302089.1714205382,
+    "generation_bytes_per_second": 315760.64445490256,
     "upload_bytes_per_second": 5284357.3788264105,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.82130391084776,
+    "estimated_full_shards_per_hour": 67.75488376841838,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -343,3 +343,4 @@
 {"step": 431400, "loss": 0.6010748744010925, "lr": 0.0003, "grad_norm": 0.14884161098757492, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63145166697, "tokens_per_second": 150872.68969278425, "data_wait_s": 1.6045996744651347e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 431500, "loss": 0.5950934290885925, "lr": 0.0003, "grad_norm": 0.15236012827299275, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63159720280, "tokens_per_second": 151064.81020809483, "data_wait_s": 1.1640004231594503e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 431600, "loss": 0.661156415939331, "lr": 0.0003, "grad_norm": 0.19191855480542128, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63174325110, "tokens_per_second": 151290.92083430433, "data_wait_s": 8.675000572111458e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 431700, "loss": 0.5878746509552002, "lr": 0.0003, "grad_norm": 0.20489848763483254, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63189055014, "tokens_per_second": 152955.85334916317, "data_wait_s": 1.7914004274643958e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:93e8bb7ac89c6a3e27318f49bc113a062ffa4193aa083258a9bc1a4f00e8b757
-size 736356
+oid sha256:5d46b767fa31bc9a7965a9a5b7294eb58d11c6bcf3241a5eee06d403f6800bff
+size 738474